---
name: distill-creator-playbook
description: 分析公开的创作者或品牌账号内容，从定位、选题、标题、开头、结构、证据、视觉、行动引导、评论和发布节奏中提炼可复用且合规的方法。适用于博主拆解、对标账号、内容蒸馏、账号分析、热门内容结构研究、创作者研究，以及把公开样本转化为原创执行手册。
---

# Distill Creator Playbook

Extract transferable decisions, not surface imitation.

## Required inputs

Establish the platform, analysis target, whether it is a benchmark or the user's own account, analysis objective, sample window, and available public evidence. If a missing choice would materially change the result, ask for it; otherwise state a reasonable assumption.

## Workflow

1. Define the question the analysis must answer: positioning, topic selection, packaging, retention, conversion, or operating cadence.
2. Build a traceable sample inventory with URL or identifier, date, format, title, topic, observable engagement, and evidence status.
3. Separate observation from inference. Mark missing, inaccessible, deleted, or login-gated material.
4. Code each item using the same dimensions: audience, job-to-be-done, hook, promise, structure, proof, emotion, visual device, CTA, and comment signal.
5. Compare high-, middle-, and low-performing samples when comparable data exists. Do not infer causation from engagement alone.
6. Identify recurring choices, controlled variations, exceptions, and possible confounders such as timing, paid distribution, celebrity, or prior audience size.
7. Distill principles at three levels: stable positioning, reusable content modules, and testable packaging tactics.
8. Translate principles into an original playbook with topic lanes, templates, examples, quality checks, and a 2–4 week test plan.
9. Identify what must not be copied: wording, proprietary assets, personal stories, trademarks, signature characters, confidential data, or unverifiable claims.
10. Return conclusions with evidence strength and unresolved questions.

## Corpus and style analysis

When the objective is to learn a writing or operating method rather than review a few posts:

1. Separate the corpus by author, platform, period, and format. Do not blend essays, short posts, ads, and formal documents into one average style.
2. Prefer complete, representative samples. Record why each item was included and keep weak or contradictory examples instead of selecting only successes.
3. Analyze six distinct layers: lexical and sentence habits; narrative structure; topic-selection logic; audience and perspective; recurring reasoning models and values; visual and layout grammar.
4. Annotate each item before making cross-sample claims. A recurring phrase is not a worldview, and a visual motif is not a content strategy.
5. Distill decision rules rather than signature wording. Preserve the creator's underlying method while producing a new voice, new examples, and new assets.
6. Test the playbook on at least one held-out sample or new brief. Record where it reproduces useful decisions, where it becomes generic, and where human judgment is still required.

## Operating-system lens

When the target exposes a repeatable content workflow, map it as a production system:

- **capture:** how observations, questions, emotions, customer language, and source material enter the system;
- **selection:** how ideas are validated against demand, account positioning, timeliness, and evidence;
- **production:** what remains human-owned and what may be assisted;
- **handoff:** the exact files, briefs, version markers, and acceptance criteria between stages;
- **approval:** decisions and external actions that require a person;
- **distribution:** how one core idea is rebuilt for different surfaces rather than copied unchanged;
- **learning:** which content and business signals update the next cycle.

Treat role names and tool counts as packaging. The transferable asset is the explicit input, output, boundary, and quality gate for each stage.

## Analysis discipline

- Use public content lawfully and respect access controls, privacy, copyright, and platform terms.
- Anonymize ordinary commenters and do not profile sensitive traits.
- Treat likes, saves, comments, and views as partial signals with different meanings.
- Do not call a pattern “爆款公式” unless repeated evidence supports it; prefer “working hypothesis.”
- Do not treat self-reported income, rankings, user counts, platform mechanisms, or causal explanations as verified merely because a post presents them confidently.
- Do not preserve tactics based on account trading, fake engagement, unauthorized copying, watermark removal, policy evasion, or manufactured controversy.
- Never invent inaccessible post content, comments, metrics, or account history.

## Route by need

- Read [references/coding-framework.md](references/coding-framework.md) when analyzing samples.
- Read [references/playbook-output.md](references/playbook-output.md) when producing the final playbook or account comparison.

## Deliverables

Return the relevant subset of: sample ledger, positioning map, topic architecture, title and opening patterns, content modules, proof and CTA patterns, visual system, comment insights, confidence notes, anti-copy boundaries, experiment backlog, and 30-day execution playbook.